Blog

Articles on AI governance, cybersecurity and systems architecture.

2026

The fridge panic and good cybersecurity practices

On 2 November 2026 a second wave of attacks on supermarket fridges reached at least six European countries. Around the same time the plasma cold rooms of the French blood service were hit, and the Pentagon is investigating similar failures at US bases. The French and British agencies explain, however, that internet-exposed controllers and weak or default passwords were enough to get in. What is known about the attack and which practices would have stopped it.

Cybersecurity
Read more

Models that read and write neural network weights directly

On 2 November 2026 a group at MIT CSAIL showed that a frontier model can read the weights of a neural network, fix them and write new ones without gradients or training. On Qwen3.5-0.8B an edit to 0.016% of the weights takes MATH-500 from 7.9% to 48.1%. What was measured, where the method stops and what it changes for model governance.

AI Governance
Read more

OccupySaaS and protest in a clean room

In mid-October 2026 laid-off developers launched OccupySaaS, which borrows the language of Occupy Wall Street and protests by using AI to build open-source replacements for the most widely used SaaS services. A protest carried out through work rather than its withdrawal, its lineage in free software and why the success of a system also depends on services, support and brand.

Open Source
Read more

The dawn of a new mathematics?

On 29 October 2026 Atoosa Kasirzadeh proposed in Noema a different way to measure progress in mathematics, now that AI models are solving the problems on the list one after another. Two models of progress, Dantzig and Wiener as examples, four fields that need mathematicians and a question that also holds outside mathematics.

AI Governance
Read more

Legal exploits

On 28 October 2026 Felix Ringe and an anonymous co-author published three vulnerabilities in German regulations found with language models: provisions that are void because the government did not cite the legal basis required by the Basic Law. How the check works, what could be exploited and which software security practices the law would need.

AI Governance
Read more

The exponentials we cannot see

On 26 October 2026 OpenAI announced an algorithm that counts the solutions of any k-CNF formula in subexponential time: if independent verification confirms it, it refutes the Exponential Time Hypothesis, on which hundreds of complexity results rest. The same announcement brings the local smoothing conjecture in every dimension and a rigorous existence proof for QCD. Exponentials remain hard for people to grasp, even though we are the product of very fast growth processes.

AI Governance
Read more

Super Intelligence

In 2014 Nick Bostrom published Superintelligence, and Elon Musk recommended it with a warning about the risks of AI. In 2026 a US executive order brings the word into federal documents, but with the 2020 definition of AI, and Musk announces that his AI division will be called SI. The superintelligence I look at is another one: made of an artificial part and a human part, built and governed with the OISG paradigm.

AI Governance
Read more

A two-speed world

On 13 October 2026 Anthropic described how it replaced git in its monorepo, which went from 2,500 to more than 30,000 merged changes a day in four months. It follows on from Intelligence is eating software: work now runs on two lanes, an agentic one and a human one, and governing the first becomes the priority of the second.

AI Governance
Read more

After supermarket fridges, more will follow

On the night of 11 to 12 October 2026 an attack on remote refrigeration controllers heated up display fridges and freezers in about 400 French supermarkets. What was hit, why the risk had been known for more than a year and why the speed at which we connect physical systems to the network widens the attack surface.

Cybersecurity
Read more

Intelligence is eating software

In 2011 Marc Andreessen wrote that software was eating the world. In 2026 an agent rebuilds an app feature starting from the binary, and within a week open rewrites of Photoshop, Word, Excel, AutoCAD and Pro Tools appear. Is software dead? The cost of writing it is falling fast, and value is moving to services, data and business functions delivered as a service.

Software Development
Read more

Building machines from code the way a cell does

On 7 October 2026 Atomic Machines presented the Matter Compiler, a system it says builds micro-machines from code alone. The idea has roots in Neal Stephenson's The Diamond Age, in the open tools that already treat hardware as code, and in the cell, which for four billion years has built its own machines, and itself, by reading instructions.

Hardware
Read more

AI and the rest of the economy

On 6 October 2026 Paul Krugman used the US national accounts to show how data-centre spending is crowding out other investment. The same signal appears in interest rates, electricity bills, memory prices and the hiring of young workers. The economic effects of AI now reach every sector, and so does the work to be done on the governance and security of the new systems.

AI Governance
Read more

Is the world we knew really over?

On 6 October 2026 OpenAI published 722 mathematics manuscripts produced by an internal model, including claimed proofs of the Unique Games Conjecture and of a weak form of the Riemann hypothesis, with Lean formalisations for 235 of 372 families. For mathematics as the production of proofs, a phase is closing. For the rest of the world, the consequences travel by slower roads.

AI Governance
Read more

AI is already building the AI that comes next

On 28 September 2026 twenty-two researchers, including Geoffrey Hinton, Yoshua Bengio, OpenAI's chief scientist and an Anthropic co-founder, published a Cambridge paper on an intelligence explosion triggered by the automation of AI research. What it argues and why the premise of its "what if" is already a measurement.

AI Governance
Read more

A new family of enzymes found by agents reading DNA

On 23 September 2026 Anthropic published the preprint of an autonomous agent campaign that swept 1.9 billion protein clusters in twenty-one and a half hours and isolated ART, a family of phage reverse transcriptases with a run of non-coding repeats beside it. How the campaign is built, how well it holds up when repeated, and what still has to be settled at the bench.

AI Governance
Read more

Will we all die?

The question has been going around for a week, between an Anthropic researcher's resignation, the extinction percentages that made it into the papers, Dario Amodei's plan for pacing the frontier and a declaration by a group of Fields medallists. My answer is no. That probability is not a measurable quantity whereas July's failures are, and the scenario that worries me is a very powerful, obedient system in the wrong hands.

AI Governance
Read more

Golden Slumbers

Listening notes on Golden Slumbers (Beatles, 1969): a lullaby from 1603 found on the music stand at home, the melody invented by someone who could not read notation, the voice straining over words that tell you to sleep.

Music
Read more